Automatic quantification of the subcellular localization of chimeric GFP protein supported by a two-level Naive Bayes classifier

Abstract

Herein we report a method supported by a two-level Naive Bayes classifier to help and improve the automatic detection and counting of cells overexpressing GFP-chimeric proteins. This toll is frequently used as a reporter for the localization and the distribution pattern of a protein in a cell. This approximation requires, besides confocal microscopy, the participation of a qualified and blind counting supervisor to avoid subjective appreciations of the imaging interpretation of the data. Indeed, this counting required specific staff training, and the interpretation of the data is inevitably subjective. In order to avoid this, we have designed an automatic detection cell counting software. We have used as a model SH-SY5Y cells overexpressing GFP-Bax protein, after 6-hydroxydopamine addition. Our proposed method learns the counting criteria after a short training stage, and uses the resulting classifier to process new images and obtaining both the number of transfected cells and the proportion of these cells that present a translocated protein. The software achieves an accuracy over 97% when detecting transfected cells, and over 93% when detecting cells with GFP-Bax translocated. Besides the hours of qualified work that can be saved, the models learnt can be stored and reused (without training) so as to homogenize criteria among different researchers.